MMarketing Against The Grain
← All frameworks
Productivity

Context-First Prompting

Feed AI the right evidence and constraints before asking it to produce work.

Difficulty
Easy
Time to result
~days to results
Steps
5
Confidence
94%

Context-First Prompting treats the prompt as only one component of an AI task. Before requesting an output, the operator assembles the relevant market data, idea details, customer evidence, target audience, competitive conditions, constraints, and desired outcome. That material is routed into the specific task, whether it is a go-to-market plan, landing page, PRD, advertisement, email funnel, or content calendar. The resulting draft is then reviewed as an editable proposal rather than accepted as authoritative. The operator checks assumptions, injects personal beliefs and domain judgment, and iterates until the recommendation is credible enough to guide action. The mechanism is evidence plus context, followed by generation, human editing, and repeated refinement.

Origin

Extracted from Marketing Against The Grain during Greg Eisenberg's explanation of how Idea Browser creates task-specific prompts.

Core principles

  • 01AI output quality depends heavily on the context supplied.
  • 02Relevant source data should travel into every downstream task.
  • 03A prompt is a working brief, not a magic command.
  • 04Generated output must be reviewed against human beliefs and evidence.
  • 05Iteration converts a generic draft into a usable deliverable.

How to run it

  1. 1

    Assemble the Evidence

    Gather the raw information the model needs, including customer quotes, market data, constraints, and prior decisions.

    Pro tip Use original source material where possible rather than summaries of summaries.

    Watch out More context is not automatically better if it is irrelevant or contradictory.

  2. 2

    Frame the Assignment

    Define the role, objective, audience, required deliverable, evaluation criteria, and practical constraints.

    Pro tip State what decision or action the output must support.

    Watch out Avoid broad requests such as asking for a complete strategy without defining the business.

  3. 3

    Route Context into the Task

    Combine the evidence and assignment into the prompt for the specific deliverable.

    Pro tip Use different task briefs for positioning, ads, a PRD, and a content calendar.

    Watch out Do not assume context from one chat or tool will automatically appear in another.

  4. 4

    Audit the Draft

    Read the output, test its assumptions, and decide which recommendations you actually believe.

    Pro tip Ask the model to identify uncertainties and alternative approaches.

    Watch out Fluent language can hide unsupported reasoning.

  5. 5

    Inject Judgment and Iterate

    Add corrections, strategic beliefs, examples, and exclusions, then regenerate or edit until the result is actionable.

    Pro tip Spend focused time in back-and-forth refinement instead of accepting the first response.

    Watch out Iteration without a clear quality criterion can become endless polishing.

In the wild

Go-to-Market Plan with Cemetery Context

A founder supplies cemetery market data, customer characteristics, pricing assumptions, competitors, and the intended lean launch. ChatGPT creates a go-to-market draft, which the founder edits to use simpler language for a nontechnical audience and to prioritize validation before a full launch.

The final strategy becomes more specific, credible, and aligned with the target customer.

Common mistakes

Prompting Without Source Data

The model fills missing context with generic assumptions that may not fit the market.

Accepting the First Draft

The initial response is a starting point and still requires review, beliefs, and iteration.

Using One Giant Generic Prompt

Different deliverables need distinct evidence, constraints, and evaluation standards.

Is it for you?

Best for

People using AI for strategy, marketing, product planning, or other tasks where context materially changes the answer.

Not ideal for

Simple factual or formatting tasks that require little situational context.

From the transcript

That data gets downloaded. We funnel that data and context into these different sections.

Greg Eisenberg · 05:00

So it's all about the context, as you know, you guys know this, right?

Greg Eisenberg · 05:00

you have to inject them into the strategy and the prompt, and you gotta iterate and go back and forth with this for 20, 30…

Host · 10:30

From the episode

How to Start a $1M Business Using Only AI